Trino

tierdb-trino is a Trino connector, install it as a plugin and point one catalog at your Postgres and your object store. It gives Trino the same consistent view the extension gives Postgres itself: the hot branch from the heap, the cold branch from the lake at the pinned snapshot, the delta overlay merged newest-wins, predicate pushdown on both branches.

Installing the plugin

Build the plugin directory and copy it into Trino's plugin directory on every coordinator and worker node:

mvn -pl connectors/tierdb-trino -am package
cp -r connectors/tierdb-trino/target/plugin/tierdb "$TRINO_HOME/plugin/tierdb"

Catalog configuration

Add a catalog properties file, etc/catalog/tierdb.properties:

connector.name=tierdb
tierdb.jdbc-url=jdbc:postgresql://your-postgres-host:5432/app
tierdb.jdbc-user=tierdb_worker
tierdb.jdbc-password=...
tierdb.fileio.s3.endpoint=https://s3.your-region.amazonaws.com
tierdb.fileio.s3.access-key-id=...
tierdb.fileio.s3.secret-access-key=...
tierdb.fileio.client.region=us-east-1

Property	Required	Meaning
tierdb.jdbc-url	yes	The same Postgres the worker and the extension use.
tierdb.jdbc-user, tierdb.jdbc-password	no	A read-only role is enough, see Roles and grants.
tierdb.pin-ttl-seconds	no	Read pin lifetime, default 900.
tierdb.fileio.*	for path-based tables	Passed through verbatim to the Iceberg S3FileIO client, stripped of the prefix (tierdb.fileio.s3.endpoint becomes s3.endpoint, and so on). Not needed when tables are catalog-based (TIERDB_CATALOG_URI set on the worker), the location is then resolved through that catalog instead.

Restart Trino, then query any registered table under the tierdb catalog like any other table:

SELECT * FROM tierdb.public.trip_events ORDER BY id;

Behavior

Reads are pinned for the lifetime of the query's split scheduling, the same tierdb.read_pins mechanism the extension and tierdb-spark use, so lake maintenance holds back snapshot expiry until the query finishes.

Predicate pushdown works on both branches: heap predicates become a WHERE clause over JDBC, lake predicates become an Iceberg filter expression. Corrections to cold rows show up immediately through the delta overlay, merged the same way as a plain Postgres read.

Column types unsupported by the connector fail loudly at query planning time rather than silently coercing, see HeapCatalog for the supported type list.

Writes are not yet supported through the connector, use the extension's SQL surface, tierdb-load, or tierdb-spark for writes, and read from Trino.